Diese Präsentation wurde erfolgreich gemeldet.
Wir verwenden Ihre LinkedIn Profilangaben und Informationen zu Ihren Aktivitäten, um Anzeigen zu personalisieren und Ihnen relevantere Inhalte anzuzeigen. Sie können Ihre Anzeigeneinstellungen jederzeit ändern.
International Journal of Technical Research and Applications e-ISSN: 2320-8163,
www.ijtra.com Volume 2, Issue 4 (July-Aug ...
International Journal of Technical Research and Applications e-ISSN: 2320-8163,
www.ijtra.com Volume 2, Issue 4 (July-Aug ...
International Journal of Technical Research and Applications e-ISSN: 2320-8163,
www.ijtra.com Volume 2, Issue 4 (July-Aug ...
International Journal of Technical Research and Applications e-ISSN: 2320-8163,
www.ijtra.com Volume 2, Issue 4 (July-Aug ...
International Journal of Technical Research and Applications e-ISSN: 2320-8163,
www.ijtra.com Volume 2, Issue 4 (July-Aug ...
Nächste SlideShare
Wird geladen in …5
×

COLOR FILTER ARRAY DEMOSAICING USING DIRECTIONAL COLOR DIFFERENCE AND GRADIENT METHOD

Nowadays digital cameras are equipped with a
single sensor (CCD/CMOS), to reduce the size and cost of the
camera. The color filter array (CFA) is used to cover the sensor
and it consist of three primary colors such as red, green and blue
and it samples only one color component at each pixel location.
The process of estimating the other two missing color components
at each pixel location is known as demosaicing. The proposed
algorithm uses the directional color difference and multiscale
gradient method for green plane interpolation, this type of
interpolation method is used to reduce the artifacts and improve
the image quality. The red and blue plane are interpolated using
the estimated green plane, the bayer pattern is used for the
interpolation technique. The performance of the image is
measured using the CPSNR value

  • Als Erste(r) kommentieren

  • Gehören Sie zu den Ersten, denen das gefällt!

COLOR FILTER ARRAY DEMOSAICING USING DIRECTIONAL COLOR DIFFERENCE AND GRADIENT METHOD

  1. 1. International Journal of Technical Research and Applications e-ISSN: 2320-8163, www.ijtra.com Volume 2, Issue 4 (July-Aug 2014), PP. 14-18 14 | P a g e COLOR FILTER ARRAY DEMOSAICING USING DIRECTIONAL COLOR DIFFERENCE AND GRADIENT METHOD R.Niruban1, Dr.T.Sree Renga Raja2, R.Deepa3 1 Electronics and Communication Engineering, Anand Institute of Higher Technology, Chennai, India 2 Electrical and Electronics Engineering, Anna University, Thiruchirapalli, India 3 Computer science and Engineering, Prince Dr.K.Vasudevan College of Engeering and Technology, Chennai, India Abstract— Nowadays digital cameras are equipped with a single sensor (CCD/CMOS), to reduce the size and cost of the camera. The color filter array (CFA) is used to cover the sensor and it consist of three primary colors such as red, green and blue and it samples only one color component at each pixel location. The process of estimating the other two missing color components at each pixel location is known as demosaicing. The proposed algorithm uses the directional color difference and multiscale gradient method for green plane interpolation, this type of interpolation method is used to reduce the artifacts and improve the image quality. The red and blue plane are interpolated using the estimated green plane, the bayer pattern is used for the interpolation technique. The performance of the image is measured using the CPSNR value. Key words— Bayer CFA, Color filter array (CFA), color peak signal-to-noise ratio (CPSNR), charged coupled device (CCD), demosaicing. I.INTRODUCTION Most digital cameras are using single sensor to capture images and it consist of single sensor with color filter array. A full color image is composed of three primary colors such as red, green and blue. The earlier digital still cameras use beam splitter which split the three primary color in three separate sensors and a color filter is placed in front of the three sensors(Fig. 1a). This approach is costly because it required three charge-coupled device(CCD) sensors. In order to reduce the cost and size of the digital camera, nowadays digital still cameras are using single charged coupled device(CCD/CMOS)with color filter array(Fig.1b). Many CFA pattern are used for demosaicing, some of them are Bayer CFA, Lukac CFA, Plataniotis CFA, Yamanaka CFA and vertical-stripe CFA are shown in Figure. 2. The most popular CFA pattern used for demosaicing is Bayer CFA, it is designed based on the luminance and chrominance level. The CFA samples only one color component at each pixel location and the color arrangement in CFA is alternative color components of red/green and green/blue. In bayer CFA the green component are placed on the quincunx lattice and red and blue component are placed on the rectangular lattice. The green color perceived brightness well because the green color sampled at a higher rate. The green color occupy most of the pixel location in CFA than the other two color because the human visual system(HVS) is more sensitive to green color than red and blue. Bayer CFA consist of 50% of green color & 25% of red and blue color. To get a full color image from the mosaiced image the missing two color component has to be interpolated at each pixel location. The process of interpolating the missing two color component is known as demosaicing. Demosaicing method is divided into two type, they are iterative method and non-iterative method. The iterative method give higher image quality when compare to non- iterative method. Many interpolation techniques are used for estimating the missing two color component. The bilinear interpolation is the simplest and most efficient method used in earlier demosaicing algorithm, but the major two artifacts are occurred in this type of interpolation techniques. (a) (b) Fig 1. Digital camera system (a) a three sensor device (b) a single sensor device. The artifacts occurs are blurring and false coloring, due to this reason many new interpolation techniques are implemented, to reduce this kind of artifacts. Later hue based interpolation is implemented is implemented it give better result in avoiding false color, when compare to bilinear interpolation but artifacts are still visible in the border region to overcome this type of interpolation technique edge behavior based interpolation method is estimated.This method give better result in image quality when compare to other interpolation technique. Another interpolation technique used to reduce artifacts is gradient based method. C. Yen Su [1] uses weighted edge- directed interpolation, this method concentrate more on the horizontal, vertical direction and edges. Lei Zhang [2] uses directional based interpolation technique to reduce artifacts. Daniele Menon [3] uses directional filtering and decision method to estimate green plane and then missing two colors are interpolated using the estimated green plane. In the proposed algorithm color difference and directionally weighted gradient based method is used to estimate the green plane and the estimated green is used to interpolate the missing red and blue plane. This interpolation method give better image quality when compare to other interpolation method discussed above. This paper is organized as follows. Section II describes the
  2. 2. International Journal of Technical Research and Applications e-ISSN: 2320-8163, www.ijtra.com Volume 2, Issue 4 (July-Aug 2014), PP. 14-18 15 | P a g e proposed methodology for determination of missing red, green and blue pixel values. Section III describes the simulation and results of the proposed algorithm. II.PROPOSED METHODOLOGY The proposed system consist of three section A) Block diagram, B) Green channel interpolation, C) Red and Blue channel interpolation. The first step of the algorithm starts with the interpolation of green plane. The color difference [and multiscale gradient method is used to interpolate the green plane. The next step in the algorithm is interpolation of red and blue plane. Then by combining these three plane the fully reconstructed. Fig. 2. Block Diagram of proposed system. color images can be obtained. Fig. 2 shows the block diagram for the proposed algorithm. A. GREEN CHANNEL INTERPOLATION Gradient are used for extracting directional data from the input images. Recently several demosaicing algorithms are make use of this method. The Green channel interpolation is based on the directionally weighted gradient method and the directional color difference [5], [14], [19] is used along with the directional data. The bayer pattern consist of red/green rows & columns and blue/green rows & columns. For blue/green rows and columns in the CFA, the directional estimates for the missing blue and green pixel values in horizontal direction are shown in equation (1). Fig 3. Bayer pattern 4 *2 2 ~ 4 *2 2 ~ 6,22,24,25,23,2 4,2 5,21,23,24,22,2 3,2 GGGBB B BBBGG G h h         (1) where h represents the horizontal direction, and represents the missing green and blue pixel values in horizontal direction. For vertical direction the missing green/blue pixel values are estimated from the equation (2). 4 *2 2 ~ 4 *2 2 ~ 3,53,13,33,43,2 3,3 3,63,23,43,53,3 3,4 GGGBB B BBBGG G v v         (2) where v represents the vertical direction, and represents the missing green and blue pixel values and (i,j) denotes the pixel location. Likewise the directional estimates for the missing red and green pixel values are estimated. Now a true value for color channel and two directional data are obtained. After estimated the directional data for missing red/green & green/blue rows and columns, then by taking their difference the directional color difference are estimated from the equation (3). )3,2()3,2( ~ )3,2( ~ , BGD h bg h  , if G is interpolated )4,2( ~ )4,2()4,2( ~ , h bg h BGD  , if B is interpolated )3,4()3,4( ~ )3,4( ~ , BGD v bg v  , if G is interpolated )3,3( ~ )3,3()3,3( ~ , v bg v BGD  , if B is interpolated (3) Here h bgD , ~ and v bgD , ~ denotes the horizontal and vertical directional color difference estimates between green and blue color. The equation explained above are similar for red/green rows and columns. The horizontal and vertical color difference are combined and calculated as follows in equation (4). 2 3,43,2 3,33,3 BB Ggb   2 3,43,6 3,53,5 BB Ggb   2 3,41,4 2,42,4 BB Ggb   2 3,43,4 4,44,4 BB Ggb   (4) Where (i,j) are pixel locations and R,G,B denotes the input values. After combining the color difference in both horizontal and vertical direction, the weight is calculated along the four direction, to calculate the weight multiscale gradient based method is used. Normally gradient based method is used to extract the directional data by using this directional data along each direction such as north, south, west and east. The weight is calculated as follows shown in equation (5), for green/blue rows and columns. 3,53,33,43,23,3 GGBBw  3,33,53,43,63,5 GGBBw  4,42,43,41,42,4 GGBBw  4,42,43,45,44,4 GGBBw  (5) Here W denotes the weight along each direction. The color difference between green/blue and weight along each direction is used to interpolate the missing green pixel vale. The estimation of green plane are calculated as follows in equation (6) & (7) )1/()(1 3,33,3 wgbk  )1/()(2 3,53,5 wgbk  )1/()(3 2,42,4 wgbk  )1/()(4 4,44,4 wgbk  )1(1()1()1/()4321( 4,42,43,53,33,4 wwwwkkkkkg  (6)
  3. 3. International Journal of Technical Research and Applications e-ISSN: 2320-8163, www.ijtra.com Volume 2, Issue 4 (July-Aug 2014), PP. 14-18 16 | P a g e The estimated green in blue CFA component is calculated as follows in equation (7) 3,43,43,4 ~ kgBG  (7) Here denotes the estimated green plane and B denotes the input blue CFA component. Similarly calculate the estimated green in red component. B. RED AND BLUE CHANNEL INTERPOLATION After interpolating the missing green plane in the red and blue pixel location, next step in the algorithm follows with the interpolation of missing red and blue component. The green plane is interpolated using the directional color difference and gradient based method, the red and blue component is interpolated using constant color difference rule and bilinear interpolation method. The red and blue plane interpolation consist of two sub-steps. The first sub-step involves the process of interpolating the missing blue component at red pixel location and missing red component at blue pixel location. The second sub-step involves the process of interpolating the missing blue and red component at green pixel location. C. MISSING RED(BLUE)COMPONENT AT BLUE(RED) PIXEL LOCATION The estimated green channel is consider along the diagonal direction to interpolate the missing red (blue) component at blue (red) pixel location. 5,63,44,52,33,41,23,4 ~~~~~~ 1 GGGGGGN  1,63,42,54,33,45,23,4 ~~~~~~ 2 GGGGGGN  (8) Here N1 and N2 denotes the normalizer, the red channel is interpolated in blue pixels location are calculated as follows using the equation (8). )3,43,4 2,52,54,34,3 3,43,4 4,54,52,32,3 3,43,4 21(*2 ) ~~ (*1 )21(*2 ) ~~ (*2~~ NN RGRGN NN RGRGN GR       (9) Here R ~ and G ~ denotes the estimated green and red component, the equation mentioned above are similar for blue channel estimation at red pixel location. D. MISSING RED AND BLUE COMPONENT AT GREEN PIXEL LOCATION The adaptive approach doesn't provide high performance gain in the interpolation of missing red and blue component in green pixel location, so the bilinear interpolation[6],[8],[16] is applied over color difference to get high performance gain. Here the closest two neighboring pixel for the original pixel value are used as shown follows in (10). 2 ) ~ () ~ (~ 2,32,32,12,1 2,22,2 RGRG GR   2 ) ~ () ~ (~ 4,34,32,32,3 3,33,3 RGRG GR   (10) Here R ~ denotes the estimated red pixel at green pixel location. Similarly missing blue plane at green pixel is calculated. After completing this step, the green, red and blue plane are combined together to get the fully reconstructed color image. This interpolation technique give better image quality and reduce artifacts. III.EXPERIMENTAL RESULTS The proposed work is tested using the 24 Kodak images (shown in fig.4) are used to test performance quality of the proposed demosaicing algorithm. The proposed algorithm is compared with some of the existing algorithm. Three existing algorithm such as bilinear interpolation [4], edge based algorithm [12], and alternate projection [10],[11] are used to compare the performance of the proposed algorithm. The performance is compared with their CPSNR values. The comparison table for the 24 Kodak image with different algorithm is shown in table I. The performance of the proposed algorithm is compared to the existing algorithms. In demosaicing the quality of the image is measured using Color Mean Squared Error (CMSE), Color Peak Signal-to- Noise Ratio (CPSNR). Color mean square error value is calculated to determine the CPSNR value. It takes the squared difference between the original image and the reconstructed image. CMSE and CPSNR are very simple techniques. CMSE involves first calculating the squared difference between the reference image and demosaiced image at each pixel and for each color channel. These are then summed and divided by three times the area of the image. Fig. 4 24 Kodak test images [15] CPSNR is then calculated using CMSE. The equations (11),(12) are shown below. CMSE CPSNR I I 2 10 255 log10 (11) Where CMSEI is a color mean square error value of the image. And its expression is given as    ji reconsorg CMSE NM jiIjiI I , 2 **3 )],(),([ (12) Where ),( jiIorg and ),( jiIrecons are the original image and reconstructed image, M and N represents the width and height of the image. The image with higher CPSNR indicates the high quality image. The proposed algorithm reduces the color artifacts and blurring effect in the image.
  4. 4. International Journal of Technical Research and Applications e-ISSN: 2320-8163, www.ijtra.com Volume 2, Issue 4 (July-Aug 2014), PP. 14-18 17 | P a g e Image Bilinear Edge based Alternative projection Proposed Algorithm 1 30.773 31.2458 37.8078 38.8294 2 36.6873 37.0425 39.6512 42.6404 3 37.6775 38.0647 41.63 43.7808 4 37.4213 37.8047 40.16 42.2015 5 31.2157 31.8233 37.5869 39.3311 6 32.1414 32.5964 38.6455 40.1151 7 37.1797 37.6152 41.0256 42.6853 8 28.2879 28.7457 35.0158 38.9985 9 36.4896 36.9292 41.2609 43.2864 10 36.3952 36.8716 40 43.2569 11 33.6645 34.1372 39.006 41.1847 12 36.9695 37.3912 40.5063 43.3886 13 28.6113 29.1087 34.2893 37.9644 14 33.469 33.9611 35.7377 40.1077 15 35.4486 35.8213 39.4739 42.5563 16 35.2366 35.6156 41.2864 41.8311 17 36.681 37.2463 40.6912 42.6004 18 32.562 33.0611 37.9911 40.0843 19 32.7829 33.2797 39.5543 41.0002 20 34.6422 34.9845 37.2988 43.1288 21 33.0627 33.5363 36.9836 41.0325 22 34.9277 35.3907 37.492 41.2275 23 38.3544 38.7374 42.7698 44.8246 24 31.411 31.8746 34.6036 40.4134 Table.1 CPSNR comparisons of proposed algorithm with existing algorithm Image Bilinear Edge based Alternative Projection Proposed Algorithm R G B R G B R G B R G B 1 29.741 34.286 29.678 29.741 39.442 29.678 38.881 40.591 39.272 37.45 51.95 37.45 2 35.022 40.797 36.100 35.022 44.854 36.100 36.524 43.533 41.665 40.90 51.95 41.20 3 35.942 41.591 37.252 35.942 45.912 37.252 42.825 45.853 42.113 42.88 51.95 41.68 4 37.106 41.126 35.663 37.106 45.051 35.663 37.928 46.396 44.927 40.53 51.95 40.65 5 30.126 34.082 30.430 30.126 40.240 30.430 37.853 41.500 37.547 37.71 51.95 37.58 6 30.803 35.699 31.355 30.803 40.606 31.355 39.616 41.921 39.609 39.70 51.95 37.47 7 35.514 41.092 36.662 35.514 46.389 36.662 42.159 45.984 40.226 42.14 51.95 40.25 8 27.104 32.181 27.220 27.104 37.959 27.220 34.998 37.820 35.697 37.30 51.95 37.31 9 36.102 40.438 34.717 36.102 45.917 34.717 42.785 45.225 42.298 41.40 51.95 42.07 10 36.272 40.017 34.528 36.272 45.509 34.528 41.980 46.187 43.416 41.47 51.95 41.92 11 32.269 37.035 33.016 32.269 41.880 33.016 39.313 43.338 41.279 40.22 51.95 38.95 12 35.266 41.164 36.404 35.266 36.404 36.404 42.376 46.138 42.872 43.23 51.95 40.77 13 27.650 31.312 27.764 27.650 35.342 27.764 37.443 38.584 36.173 36.37 51.95 36.14 14 32.123 36.675 32.828 32.123 41.524 32.828 34.990 40.838 36.262 39.02 51.95 37.92 15 34.216 39.315 34.439 34.216 43.306 34.439 37.339 43.381 41.055 40.39 51.95 41.62 16 33.815 39.277 34.378 33.815 43.662 34.378 43.143 44.836 42.260 41.98 51.95 38.95 17 36.205 39.248 35.440 36.205 43.949 35.440 43.318 45.433 42.754 40.99 51.95 41.02 18 31.906 35.128 31.478 31.906 38.980 31.478 37.865 41.464 38.428 38.41 51.95 38.41 19 31.592 36.521 31.772 31.592 42.845 31.772 39.476 42.313 40.337 39.53 51.95 39.18 20 33.838 39.032 33.080 33.838 43.287 33.080 41.613 44.279 39.349 41.99 51.97 41.16 21 31.892 36.279 32.221 31.892 40.838 32.221 40.720 42.780 39.923 40.07 51.95 38.80 22 33.866 38.111 33.983 33.866 42.446 33.983 38.562 42.531 38.472 39.45 51.95 39.73 23 36.114 42.556 38.598 36.114 47.324 38.598 41.226 46.550 42.145 43.12 51.95 43.59 24 31.062 34.104 30.016 31.062 37.717 30.016 37.292 40.023 36.291 38.57 51.94 38.94 TABLE.2 PSNR Comparisons of Proposed Algorithm with Existing Algorithm The table I and II shows that the CPSNR and PSNR value of bilinear interpolation is less when compare to proposed method because it doesn't consider the edges, so the artifacts are produce along the edges. The edge based method used edge based interpolation only for the green channel interpolation, so the CPSNR and PSNR value is less when compare to proposed method. The alternative projection method used transform only for red and blue channel, so the CPSNR and PSNR value is less when compare to proposed method. The proposed method give better image quality when compare to above mentioned algorithm, because directional color difference and gradient based method is used for green channel interpolation and the estimated green plane is used for red and blue plane interpolation. IV.CONCLUSION The color artifacts and false colors are some common artifacts in the demosaicing algorithm. The proposed method used directional color difference and gradient based method to reduce such artifacts. This method give better image quality and the computational complexity of this technique is less. CPSNR value shows that the proposed algorithm is able to produce better demosaicing results, when compare to other algorithms. This shows that the proposed method can prove to be useful for image processing problems. REFERENCES [1] Yen Su, C, Highly iterative Demosaicing using Weighted- Edge and Color-Difference Interpolations. IEEE Transactions on Consumer Electronics, vol. 52, no. 2, pp: 639-646, 2006. [2] Menon, D, Andriani, S, and Calvagno, G, Demosaicing with directional filtering and a posteriori decision. IEEE Transactions on Image Processing, vol. 16, no. 1, pp: 132–141, 2007. [3] Zhang, L, and Wu, X, Color demosaicking via directional linear minimum mean square-error estimation. IEEE Transactions on Image Processing, vol. 14, no. 12, pp: 2167-2178, 2005. [4] Vinsley, S.S, Adaptive weighted demosicing for single sensor camera images. International journal of imaging science and engineering, vol. 2, no. 2, pp: 201-206, 2008. [5] Chung, K.H, and Chan, Y.H, Color demosaicing using variance of color difference. IEEE Transactions on Image Processing, Vol. 15, no.10, pp: 2944-2955, 2006. [6] Kimmel, R, Demosaicing: image reconstruction from CCD samples. IEEE Transactions on Image Processing, vol. 8, pp: 1221-1228, 1999.
  5. 5. International Journal of Technical Research and Applications e-ISSN: 2320-8163, www.ijtra.com Volume 2, Issue 4 (July-Aug 2014), PP. 14-18 18 | P a g e [7] Hur, B.S., and Kang, M.G, High definition color interpolation scheme for progressive scan CCD image sensor. IEEE Transactions on Image Processing, vol. 47, no. 1, pp: 179-186, 2001. [8] Li, X, Demosaicing by Successive Approximation. IEEE Transactions on Image Processing, vol. 14, no. 3, pp: 370-379, 2005. [9] Hamilton, J.F, and Adams, J.E, Adaptive color plane interpolation in single sensor color electronic camera. U.S. Patent 5 629 734, 1997. [10] Laroche, C.A, and Prescott, M.A, Apparatus and method for adaptively interpolating a full color image utilizing chrominance gradients. U.S. Patent 5 373 322, 1994. [11] Gunturk, B.K, Altunbasak, Y, and Mersereau, R.M, Color plane interpolation using alternating projections. IEEE Transactions on Image Processing, vol. 11, no. 9, pp: 997–1013, 2002. [12] Paliy, D, Katkovnik, V, Bilcu, R, Alenius, S., and Egiazarian, K, Spatially adaptive color filter array interpolation for noiseless and noisy data. Int. J. Imag. Syst. Technol, pp: 105–122, 2007. [13] Li, X, Gunturk, B, and Zhang, L, Image demosaicing: A systematic survey. Proc. SPIE–Int. Soc. Opt. En., 68221J-1-15, 2008. [14] Hirakawa, K, and Parks, T.W, Adaptive homogeneity-directed demosaicing algorithm. IEEE Transactions on Image Processing, vol. 14, no. 3, pp: 360–369, 2005. [15] Kehtarnavaz, N, Oh, H.J, and Yoo, Y, Color filter array interpolation using color correlations and directional derivatives. Electronic Imaging Journal, 621-632, 2003. [16] Pei, S.C, and Tam, I.K, Effective color interpolation in CCD color filter arrays using signal correlation. IEEE Transactios on Circuits and Systems for Video Technology, vol. 13, no. 6, pp: 503–513, 2003. [17] Kakarala, R, and Baharav, Z, Adaptive demosaicing with the principle vector method. IEEE Transactions on Consumer Electronics, vol. 48, pp: 932–937, 2002. [18] Lukac, R, Platoniotis, K.N, Data adaptive Filters for Demosaicing: A Framework. IEEE Transactions on Consumer Electronics, vol. 51, pp: 560–570, 2004. [19] Pekkucuksen, I, and Altunbasak, Y, Edge strength based color filter array interpolation. IEEE Transactions on Image Processing, vol. 21, no. 1, pp: 393–397, 2009. [20] Muresan, D.D, and Parks, T.W, Demosaicing using optimal recovery. IEEE Transactions on Image Processing, vol. 21, no. 1, pp: 267–278, 2005. R.Niruban was born in Tamilnadu, India, in 1984.He received the B.E. degree (with first class with distinction) in Electronics and Communication Engineering in 2005, the M.E. degree (with first class) in Applied Electronics in 2007,from Anna University, Chennai. He is currently working as Assistant Professor in Electronics and Communication Engineering department. His research interests are in Digital Image Processing and their applications including topics in Demosaicing, Image Inpainting, and Histogram processing.

×